The present invention relates to a method and equipment for detecting a pattern defect; and, more specifically, the invention relates to a method and equipment suitable to detect and test a defect of a pattern formed in a semiconductor wafer, a liquid crystal display, a photomask, etc.
Conventionally, detecting equipment, such as described in Japanese Published Unexamined Patent Application No. 7-318326 (prior art No. 1), scans an image of a pattern under test (hereinafter referred to as a “test pattern” for simplicity) using an imager, such as a line sensor etc., and is able to recognized nonconformity as a defect by comparing grayscale levels of the detected image signal to an image signal delayed by a prescribed time while moving the test pattern.
Moreover, a conventional technology concerning detection of a defect in a test pattern is disclosed in Japanese Published Unexamined Patent Application No. 8-320294 (prior art No. 2). This prior art 2 is intended to be applied to a test pattern on a semiconductor wafer etc. where a high density area of the test pattern, such as a memory mat part etc., and a low density area of the test pattern, such as a peripheral circuit etc., exist in a mixed manner. This publication describes a method comprising the steps of: converting a digital image signal that is obtained through AD conversion of an image signal detected from the above-described test pattern into grayscale levels so that the brightness or the contrast ranging between the high density area and the low density area of the test pattern is converted into a predetermined relation based on a brightness-frequency relationship of the above-described detected image signal; performing a function approximation on both the image signal thus grayscale converted and an image signal to be compared (hereinafter referred to as a “comparison image signal”) therewith, which was also grayscale converted; integrating the difference between the two curves represented by the function approximations; aligning two image signals which were grayscale converted based on information of high-precision detection of misalignment obtained from the integral value; and detecting a minute defect with high-precision by comparing test patterns while keeping the alignment between two image signals optimally.
Moreover, in the case of detecting a photomask, conventionally it has been proposed that the light used in the detecting should be the same as the exposure light so as to detect only a detrimental defect which will cause trouble in the actual exposure; accordingly, with this in mind, it has been suggested that inspection of a photomask exposed with ultraviolet light (hereinafter referred to as “UV light”) should be performed using the same UV light as the exposure light. Patent applications concerning this technology, as a technology to test the appearance of a circuit pattern on a photomask, include Japanese Published Unexamined Patent Application No. 8-94338 (prior art No. 3) and No. 10-78668 (prior art No. 4).
In addition, a technology to measure the amount of phase shift in a phase shift mask is disclosed in Japanese Published Unexamined Patent Application No. 10-62258 (prior art No. 5) and No. 10-78648 (prior art No. 6).
Furthermore, a technology to clearly visualize a circuit pattern and a foreign material optically by inspecting a specimen with visible light and UV light by making good use of a fact that materials used in a process have different absorption characteristics for visible light and UV light is disclosed in Japanese Published Unexamined Patent Application No. 4-165641 (prior art No. 7) and No. 4-282441 (prior art No. 8).
Moreover, means for measuring optically an external form of an object using an interferometer is disclosed in Japanese Published Unexamined Patent Application No. 4-357407 (prior art No. 9), wherein UV light is applied to the interferometer.
LSI fabrication in recent years has progressed toward finer microfabrication in circuit patterns formed on wafers in response to a need for high-integration, and a pattern having a width (feature size) as small as 0.25 μm or less is being required, reaching almost a limit of the available imaging optical systems. Therefore, efforts to attain a high NA in an imaging optical system and to apply the optical super-resolution technology, as well as efforts to provide more sophisticated image processing, are being made. The above-described prior arts 1 and 2 are directed to techniques that use those results. However, implementation of a high NA has already reached its physical limit, and this measure has a problem of weakness for patterns having a large pattern step height. Also, the optical super-resolution technology and sophistication of image processing have a problem of limited applicability because of their non-linear response.
Therefore, an attempt to shorten the wavelength of light used in defect detection, from a visible radiation region in conventional use to a UV light region, is an essential approach.
On the other hand, the idea that the same light source as exposure light should be used, which has been originally devised for a photomask, is effective for prior arts 5 and 6 for measuring the amount of phase shift. This is because the amount of phase shift is directly linked with the wavelength of the light source. However, in case defects are to be detected by detecting the appearance of the whole surface of a test sample or a large area of a circuit pattern comparable to it, the technology wherein a wavelength of detecting light is chosen to be the same as the exposure light (prior arts 3 and 4) is not necessarily an appropriate technique.
This is because the pattern transfer capability by exposure cannot be determined only by the wavelength of the light source and the conditions of the optical system. The transfer capability is closely connected with various factors in a complicated way, such as the amount of exposure, properties of a resist, the amount of defocusing, an optical characteristic of an underlying material, a developing process, etc. Consequently, the prior arts 3 and 4 are directed to techniques which are suitable to analyze carefully the pattern transfer capability of a single defect by performing a simulation including these complicated conditions, but are different from a technology for detecting defects of a large number of circuit patterns in a short period of time.
In the case where a large number of circuit patterns are examined in a short time, it will be a practical solution for this problem to thoroughly detect any defects having a possibility of being transferred as a detectable defect with a sensitivity as high as possible by means of a light source that is chosen only to detect defects, rather than performing a detection by applying an expensive, hard-to-handle exposure light source.
In this case, since UV light is employed to improve the resolution, visible light that deteriorates the resolution cannot be employed jointly as is the case of the prior arts 7 and 8.
Further, since it is essential to perform a rapid detecting, a minutely converged laser beam as in the prior art 9 cannot be used. In the UV light region, since a high-illuminance discharge lamp does not exist, a high-illuminance illumination by means of a laser is indispensable. However, as a result, when a laser beam is expanded to a whole field of view, an interference fringe pattern due to interference of the laser beam, a so-called speckle pattern, occurs and overshoot and undershoot occur in edge-portions of a circuit pattern, which make it impossible to obtain images.
Laser beams have excellent features as light sources. To use them in a way which will give their features full play, when a certain area is illuminated, generally the laser beams are scanned using some kind of scanning means.
For the scanning means, there are means capable of scanning by driving a mirror mechanically to change a reflection direction, means capable of scanning by applying an electric signal to an optical crystal to effect a change in diffraction direction or in refraction direction, and the like.
Among the former means, there exist a galvano mirror, a polygon mirror (a polyhedron mirror), etc., and among the latter means, there exist an A/O deflector, an E/O deflector, etc.
Japanese Published Unexamined Patent Application No. 7-201703 discloses equipment to scan a laser beam using a polygon mirror or a galvano mirror and to write a pattern with a minutely converged laser spot. Further, Japanese Published Unexamined Patent Application No. 8-15630 discloses equipment to scan a laser beam using a polygon mirror or an A/O deflector write to a pattern with a minutely converged laser spot. Furthermore, Japanese Published Unexamined Patent Application No. 10-142538 discloses equipment to write a pattern with a minutely converged laser spot in a scheme where two polygon mirrors are set in synchronization with each other, a phase difference of half the period between mirror facets being set, to perform the scanning, and further these polygon mirrors are switched between one another, so that a combination of polygon mirrors scans a laser beam with improved efficiency. Furthermore, Japanese Published Unexamined Patent Application No. 7-197011 discloses a device wherein two polygon mirrors are stacked up with a phase difference of half the period between mirror facets thereof being set and a semiconductor laser diode is modulated in synchronization with its rotation. Furthermore, Japanese Published Unexamined Patent Application No. 5-34621 discloses a device wherein mirror facet angles of a polygon mirror are varied from facet to facet, so that two-dimensional beam scanning can be performed.
The scanning by a polygon mirror (polyhedron mirror) has a problem in that since the scanning is performed under continuous rotation, the scanning is unavailable at the edges of mirror facets, the effective scanning time is decreased, and as a result a decrease inefficiency is brought about.
Moreover, if high-speed scanning is intended, a plurality of polygon mirrors cannot be rotated in synchronization with each other because of the continuous rotation. Therefore, it is impossible in a high-speed region for a two-dimensional area to be scanned by integrating polygon mirrors, and for the scanning efficiency of polygon mirrors to be improved in a scheme where two polygon mirrors are used to perform the scanning in synchronization with each other, also with a phase difference of half the period between mirror facets thereof being set, and which are switched over for use, as disclosed in Japanese Published Unexamined Patent Application No. 10-142538.
Further, a method for directly modulating a laser, even in a scheme where two polygon mirrors are stacked up with a phase difference of half the period between mirror facets thereof being set, as disclosed in Japanese Published Unexamined Patent Application No. 7-197011, is not suitable for application in a deep ultraviolet wavelength region, because gas lasers and solid state lasers are not suitable for direct modulation. Further, one such laser is too expensive to make a configuration where a plurality of lasers are arranged and on-off switched instead of being directly modulated.
Further, since a polygon mirror rotates continuously, the scanning orange cannot be changed. Therefore, the shape of the range scanned by a polygon mirror is limited to a rectangular area and hence polygon mirrors are not suitable to scan circular regions.
Further, in the wavelength region of deep ultraviolet light sources, there is a problem that a surface irregularity of a polygon mirror causes scattered light, which deteriorates the beam quality and refection efficiency.
As for the scanning by galvano mirrors, ordinary galvano mirrors can only perform low-speed scanning with a scanning speed of a few hundred Hz at maximum, whereas a resonant-type galvano mirror can perform high-speed scanning at a few kHz, but the driving signal is limited only to sinusoidal waves and the scanning angle varies sinusoidally, and therefore the scanning speed of the beam is not constant. Because of this fact, when laser beam scanning is performed for illumination to obtain a detected signal especially using a storage-type sensor, there is a problem in that a signal from a slowly scanned area is relatively large, whereas a signal from a fast scanned area is relatively small.
For E/O deflectors, there is no crystal usable in a wavelength range of ultraviolet to deep ultraviolet light. Therefore, current technology cannot respond to a request for a high-resolution optical system with a light source whose wavelength is shortened.
Moreover, for A/O deflectors there is only quartz for a crystal usable in the range of ultraviolet to deep ultraviolet wavelengths. However, the acoustic velocity in quartz is large. This fact is not a large obstacle when an A/O element is used as a modulator, but becomes a problem when it is used as a deflector. When a diffraction grating is formed in quartz using an acoustic element, a variation of the spacing between the gratings in response to a change of signal frequency applied to the acoustic element is small because of its large acoustic speed.
This means that an angular region in which a deflector can deflect light is small. Since the acoustic velocity in quartz is about 6000 m/s and an upper limit of the signal frequency that can be applied to an acoustic element is around 150 MHz, deflection of only 0.23 degree is achievable provided that the variation range of frequency is ±100 MHz. Therefore, if a sufficient scanning range is intended to be achieved, an extremely long optical path (1 m or more) should be provided. With provision of a long optical path, there arises a problem of deterioration in beam position and beam quality due to an environmental change, such as fluctuation of air in the optical path.